System and method for dynamic image recognition

A method for dynamic image recognition for making determinations about regions of interest on an object using an image recognition computer program contained in computer-readable form on a tangible storage medium, comprising the steps of executing the image recognition computer program to:

• acquire raw gray level image of the object comprising a set of raw pixels, eac raw pixel having a gray level value;
• derive a number of distinct spaces from the raw pixel image;
• cluster the raw pixel images for each distinct space based on a set of clustering rules;
• and
• generate a set of generated features for each region of interest in each distinct space, each generated feature having an arbitrary geometry;
• load a knowledge base containing a set of presence/absence features, polarity features and past decision confidences for each region of interest for each derived space for the object to be imaged;
• score each generated feature for each region of interest on each of the derived spaces;
• classify each generated feature for presence/absence or polarity;
• calculate decision confidence for each generated feature;
• and
• calculate a relative decision confidence for each generated feature to provide a metric that compares the decision confidence for each generated feature to past decision confidences for the corresponding feature contained in the knowledge base;
• if the relative decision confidence is outside of a predetermined window, classify the generated feature as a defect;
• and
• perform an extended relative confidence analysis to subclassify the defect;
• and
• create the knowledge base by:
• generating each presence/absence feature for each derived space from at least one blank object;
• scoring each presence/absence feature for each derived space from a first blank object;
• and
• scoring each presence/absence feature for each derived space from every other blank object imaged;
• generating each polarity feature for each derived space from a first assembled object;
• rotating the feature images to generate wrong polarity features;
• and
• scoring each polarity feature for each derived space for every assembled object imaged.

A system of dynamic image recognition for making determinations about regions of interest on an object, comprising:

• a processor;
• and
• an image recognition computer program stored in computer-readable form on a tangible storage medium, the image recognition computer program executable to:
• acquire a raw gray level image of the object comprising a set of raw pixels, each raw pixel having a gray level value;
• derive a reduced zero order magnitude space;
• derive a reduced first order magnitude space;
• derive a first order direction space;
• derive a reduced second order magnitude space;
• derive a second order direction space.
• cluster the raw pixel images for each distinct based on a set of clustering rules;
• generate a set of generated features for each region of interest in each distinct space, each feature having an arbitrary geometry;
• load a knowledge base containing a set of stored features and past decision confidences for each region of interest for each derived space for the object to be imaged;
• score each generated feature for each region of interest on each of the derived spaces;
• classify each generated feature for both presence/absence and polarity;
• calculate decision confidence for each generated feature;
• and
• calculate a relative decision confidence for each generated feature to provide a metric that compares the decision confidence for each generated feature to past decision confidences for the corresponding stored feature contained in the knowledge base;
• and
• if the relative decision confidence is outside of a predetermined window, classify the generated feature as a defect and perform an extended relative confidence analysis to subclassify the defect.

The system for dynamic image recognition for making determinations about regions of interest on an object, comprising;

• a processor;
• and
• an image recognition computer program stored in computer-readable form on a tangible storage medium, the image recognition computer program executable to:
• acquire a raw gray level image of the object comprising a set of raw pixels, each raw pixel having a gray level value;
• derive a number of distinct spaces from the raw pixel image;
• cluster the raw pixel images for each distinct space based on a set of cluster rules;
• and
• generate a set of generated features for each region of interest in each distinct space, each generated feature having an arbitrary geometry;
• load a knowledge base containing a set of presence/absence features, polarity features and past decision confidences for each region of interest for each derived space for the object to be imaged;
• score each generated feature for each region of interest on each of the derived spaces;
• classify each generated feature for both presence/absence and polarity;
• calculate decision confidence for each generated feature;
• and
• calculate a relative decision confidence for each generated feature to provide a metric that compares the decision confidence for each generated feature to past decision confidences for the corresponding feature contained in the knowledge base;
• and
• if the relative decision confidence is outside of a predetermined window, classify the generated feature as a defect and perform an extended relative confidence analysis to subclassify the defect;
• and
• wherein the image recognition software is further executable to create the knowledge base by:
• generating each presence/absence feature for each derived space from at least one blank object;
• scoring each presence/absence feature for each derived space from a first blank object;
• and
• scoring each presence/absence feature for each derived space from every other blank object and each assembled object imaged;
• generating each polarity feature for each derived space from a first assembled object;
• rotating the feature images to generate wrong polarity features;
• and
• scoring each polarity feature for each derived space for every assembled object imaged.